app.py

from fastapi import FastAPI, HTTPException
from fastapi.responses import FileResponse
from fastapi.middleware.cors import CORSMiddleware
import uvicorn
import os
import tempfile
import pickle
from vinorm import TTSnorm
from f5_tts.model import DiT
from f5_tts.infer.utils_infer import load_vocoder, load_model, infer_process
from huggingface_hub import hf_hub_download, snapshot_download
import soundfile as sf

# Load models VÀ voice cùng lúc
hf_token = os.environ.get("HF_TOKEN")

print("🔄 Đang tải models và voice...")

# 1. Load TTS model
vocoder = load_vocoder()
model_ckpt = hf_hub_download(repo_id="GexSay/stt1beta", filename="model_last.pt", repo_type="model", token=hf_token)
vocab_file = hf_hub_download(repo_id="GexSay/stt1beta", filename="config.json", repo_type="model", token=hf_token)
model = load_model(DiT, dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4), ckpt_path=model_ckpt, vocab_file=vocab_file)

pkl_dict = {}

app = FastAPI(title="Bankme TTS API")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

def post_process(text: str):
    text = " " + text + " "
    text = text.replace(" . . ", " . ").replace(" .. ", " . ")
    text = text.replace(" , , ", " , ").replace(" ,, ", " , ")
    text = text.replace('"', "")
    return " ".join(text.split())

@app.get("/")
async def root():
    return {"message": "Bankme TTS", "status": "running"}

@app.post("/tts")
async def generate_tts(voice: str, text: str, speed: float = 1.0):
    try:
        # Validate input
        if not voice:
            raise HTTPException(status_code=400, detail="Voice is required")
        if not text.strip():
            raise HTTPException(status_code=400, detail="Text is required")
        
        if voice in pkl_dict:
            pkl_path = pkl_dict[voice]
        else:
            print(f"🔄 Voice '{voice}' chưa có local, thử tải từ HF Hub...")
            try:
                pkl_path = hf_hub_download(
                    repo_id="GexSay/stt1beta",
                    filename=f"voice/{voice}.pkl",
                    repo_type="model",
                    token=hf_token
                )
                pkl_dict[voice] = pkl_path
                print(f"✅ Đã tải voice '{voice}' thành công")
            except Exception as e:
                print(f"❌ Không thể tải voice '{voice}' từ HF Hub: {e}")
                raise HTTPException(
                    status_code=404, 
                    detail=f"Voice '{voice}' not found and cannot be downloaded. Available voice: {available_voice}"
                )

        # Load voice data từ pickle
        with open(pkl_path, "rb") as f:
            audio, sr, ref_text = pickle.load(f)
        
        # Process text
        processed_text = post_process(TTSnorm(text, punc=True)).lower()
        
        # Generate audio
        final_wave, final_sr, _ = infer_process(
            audio, sr, 
            ref_text.lower(), 
            processed_text, 
            model,
            vocoder,
            nfe_step=8,
            speed=speed
        )
        
        # Save to temporary file
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_file:
            sf.write(tmp_file.name, final_wave, final_sr)
            temp_path = tmp_file.name
        
        return FileResponse(
            temp_path,
            media_type="audio/wav", 
            filename=f"tts_{voice}.wav"
        )
        
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Generation error: {str(e)}")

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=7860)